- Try more complex generator
- ...so simpler the discriminator
- A "stronger" dataset: the style of the distribution should be more highlighted
Notes about training GAN
G' and G''
Exercises 7.41 and 8.14 from Rubinstein's book Polymer Physics.
From Boltzmann superposition principle, we have
We therefore have
and represent Fourier cosine and sine transform. Take and (note the normalization factor of ), let :
and are defined as the Fourier sine and cosine transform of stress relaxation function :
Let be
G' i.e., the cosine transform is thus
Since , we have (let )
Since , extending the range of last integral to yields a Gaussian integral with complex argument, the result is straightforward:
With and , we finally have
The factor comes from the definition of Fourier cosine transform.
P.S. It is more convinient (non-restrictively) to treat the integral
with as Gamma function with (complex) coefficient in :
We may use Hypergeometric function for more genernal cases, i.e.,
P.S. It is more convinient (non-restrictively) to treat the integral
Subscribe to:
Posts
(
Atom
)